mcg-nju / mgsampler Goto Github PK

I have same questions,After i took a earnest look at your code .how is the JSON file generated.i would appreciate your reply.

Thank you for your great work.
My question is that why you use the function skimage.measure.compare_ssim to compute the similarity of two adjacent frames instead of subtracting directly as your paper says.

Is there a trained model could be provided. I just want to do a test. Thanks.

Hello, I reported these two problems after running the script file, what is the reason？

KeyError: 'SampleFrames is already registered in pipeline'
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 1525602) of binary: /home/zhouyx/anaconda3/envs/open-mmlab/bin/python

Hi! I have a doubt when I inserted MGSampler into the original codes. What are the img_diff_sthv1_train.json and img_diff_sthv1_val.json in ./MGSampler/mmaction/datasets? How can I get them?

with open('/home/zhiyuan/img_diff_sthv1_train.json', 'r') as f:
img_train = json.load(f)
with open('/home/zhiyuan/img_diff_sthv1_val.json', 'r') as f:
img_val = json.load(f)

Thank you!